Optimization method for fusing network scene edge data transmission and calculation task unloading

Abstract

The invention provides an optimization method for fusing network scene edge data transmission and calculation task unloading. Through the steps of protocol identification, local protocol conversion, cloud-assisted unknown protocol processing, data packet tunnel packaging, data packet analysis and storage, state transition processing and the like, various communication protocol data packets are effectively processed, and the data transmission efficiency is optimized. In addition, according to the technical scheme, by establishing a comprehensive model and adopting an improved depth deterministic strategy gradient algorithm, task unloading and migration decisions are dynamically evaluated and optimized, delay is reduced, and the processing speed and quality are improved. The invention provides an efficient, flexible and multi-protocol compatible optimization method for overcoming the defects of an existing edge computing solution in the aspect of collaborative optimization of data transmission and task processing, and the network performance and the user experience are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of communication networks and data processing technologies, and relates to a collaborative optimization method for data transmission and edge computing task offloading. Background Art

[0002] With the rapid development of advanced technologies such as cloud computing, the Internet of Things, and 5G, data transmission and computing needs have exploded in various industries. At present, the network has increasing requirements for data transmission efficiency and computing task processing capabilities. In this context, edge computing, as an emerging computing paradigm, aims to reduce network latency, increase data processing speed, and optimize user experience by processing data at the edge of the network near the user end. However, existing edge computing solutions still face many challenges in handling multi-protocol data transmission, computing task offloading, and coordinated optimization with network transmission.

[0003] On the one hand, there are many communication protocols in the existing network environment, such as TCP, LoRa, RS232, etc. The poor compatibility between these protocols makes the transmission and processing of data packets complicated and inefficient. In addition, with the surge in the number of devices, the types and number of data packets are also increasing rapidly, which has brought tremendous pressure to network transmission and computing task processing. On the other hand, task offloading strategy is crucial in edge computing. Reasonable task offloading can not only reduce computing delays, but also effectively utilize edge computing resources. However, how to dynamically determine task offloading based on the real-time network status and the current load of edge nodes, and how to perform task migration under the changes in the network environment caused by user mobility, is still a technical problem that needs to be solved urgently.

[0004] In summary, with the continuous expansion of edge computing application scenarios, higher requirements are placed on the coordinated optimization of data transmission and computing task processing. How to ensure efficient processing of computing tasks while ensuring data transmission efficiency has become a key technical challenge in edge computing. Therefore, there is an urgent need for a technical solution that can effectively solve the above problems to support the coordinated optimization of efficient data transmission and computing task offloading. Summary of the invention

[0005] In view of the above technical problems, the present invention proposes an optimization method for edge data transmission and computing task offloading in a computing network scenario, including:

[0006] Step (1): Perform protocol identification (including TCP, UDP, RS232, RS485, CAN, Modbus and other communication protocols) and preliminary processing. After receiving the data packet, the switch extracts and parses the packet header field to determine whether the data packet is a locally known protocol, and completes the identification of the data transmission destination address, source address, transmission type and other status parameters. For example, in a multi-protocol industrial network, due to the differences in frame structure and data transmission characteristics between RS485 and CAN protocols, traditional equipment is difficult to achieve efficient compatibility in a multi-protocol environment, which can easily lead to data transmission delays or losses. Through this step, the switch can quickly identify and classify data packets, ensure accurate processing of each protocol data, avoid conflicts between protocols, thereby improving the compatibility and data processing efficiency of the multi-protocol network, and providing a reliable communication foundation for complex network environments.

[0007] Step (2): Perform local protocol translation and conversion. For data packets of locally known protocols, the switch starts the protocol conversion mechanism to complete the translation process based on the protocol type and forwarding port information. The specific steps include: filling in the necessary protocol data, rebuilding and replacing the original data packet header, and storing the original data packet header information in a standardized manner. This process ensures that the switch can effectively perform conversion operations of different protocol formats in a multi-protocol environment, thereby achieving accurate transmission of data in a heterogeneous network environment. For example, in the intercommunication scenario of TCP and LoRa protocols, the switch seamlessly completes the adaptation and interoperability between protocols through a dynamic protocol conversion mechanism, successfully solving the problem of poor protocol compatibility in heterogeneous networks, and significantly improving data transmission efficiency and network throughput.

[0008] Step (3): The cloud assists in processing unknown protocols. If the data packet is not a locally known protocol, the switch sends a query or processing request to the cloud server, including the protocol information and transmission information of the data packet. This mechanism significantly enhances the compatibility and adaptability of the system in a heterogeneous protocol environment by leveraging the powerful computing power and knowledge base resources of the cloud. For example, in the edge computing scenario of the industrial Internet of Things, when sensors use non-standard protocols, the switch implements protocol translation through cloud parsing, so that the data packet can be accurately transmitted to the central control node, thereby ensuring the integrity and validity of the data flow. If the cloud fails to parse the protocol, the switch records the relevant error information and feeds it back to the knowledge base to support subsequent system optimization and protocol expansion. This method can effectively solve the interoperability problem in a multi-protocol environment and improve the system's dynamic adaptation capability and data transmission efficiency.

[0009] Step (4): Packet tunnel encapsulation. After receiving the data packet, the switch first parses the packet header field and queries the local mapping information. Based on the network transmission protocol, the switch constructs and encapsulates a new protocol header, sets a new destination, sending address and port to ensure that the data can be accurately transmitted to the target node. If the local mapping query fails, the switch sends a request to the cloud server, which queries the mapping library and returns control information to guide the switch to make necessary adjustments. The switch updates the mapping table based on the cloud feedback, modifies the packet field, and completes the final encapsulation. This process effectively supports cross-protocol data transmission, especially in dynamic network environments and complex protocol conditions, and can improve the adaptability of the system, the reliability and effectiveness of data transmission. For example, when a data packet is converted between the TCP protocol and the LoRa protocol, the switch can flexibly handle protocol differences to ensure that the data can be seamlessly transmitted to the target node, significantly optimizing the utilization of network resources.

[0010] The purpose of the above four steps is to align the formats of multiple data packets from multiple communication networks. The following will explain the data packet reception processing and mapping to the Data Plane Development Kit (DPDK) function library for subsequent high-speed forwarding.

[0011] Step (5): Data packet parsing, storage and state transition processing. The Ethernet header is parsed by the parse_ethernet function to extract the target MAC address, source MAC address, Ethernet type and check sequence. The state jump is determined according to the etherType field value: 0x9999 to network identification resolution, 0x0806 to ARP, 0x0800 to IPv4, 0x86DD to IPv6, and other values ​​are accepted by default. The parse_arp, parse_sinet, parse_ipv4, and parse_ipv6 functions process the corresponding protocol headers respectively, directly receiving or deciding further state jumps. Among them, parse_arp parses the ARP packet, extracts the operation code, source / destination MAC and IP address, and decides to directly receive or perform address resolution; parse_sinet parses the SINet packet, extracts the protocol identifier and data priority, and performs priority processing or standard load forwarding; parse_ipv4 and parse_ipv6 process the IPv4 / IPv6 packet respectively, extract the source / destination address and protocol field, and jump to the fragment reassembly state if there is fragmentation to ensure the correct reception and processing of the data packet. For example, in a multi-protocol environment, when a device receives data packets of TCP and LoRa protocols, traditional processing methods may lead to poor protocol compatibility, transmission delays, and inefficiency. Through the parsing and state migration mechanism, the switch uses the etherType field to quickly identify the protocol type and select the most appropriate processing flow. For ARP packets, the parse_arp function determines whether address resolution is required based on the opcode and MAC / IP address to avoid delays; the parse_sinet function adjusts the forwarding strategy based on the priority to ensure timely processing of high-priority data; for IPv4 / IPv6 packets, if fragmentation is detected, the system jumps to the fragment reassembly state to ensure data integrity. This dynamically adjusted processing flow can solve protocol compatibility issues, improve data processing efficiency and network throughput, and optimize task offloading and data transmission efficiency in edge computing.

[0012] Step (6): Perform mapping lookup. Mapping lookup involves six processing strategies: ARP, SINET, IPv4, IPv6 processing, and SINET encapsulation and decapsulation. Executing these strategies according to specific parameters can effectively solve compatibility issues in multi-protocol environments. First, the ARP strategy matches the target IP address, searches for the target MAC address in the local ARP table, and fills in the source MAC and IP addresses to ensure that the data packet can be accurately forwarded and avoid delays and packet losses caused by address mismatches. For the SINET protocol, the switch determines the MAC address of the next hop based on the destination identifier and updates the Ethernet frame header to ensure smooth data packet transmission and avoid forwarding failures caused by target address mismatches. The IPv4 / IPv6 strategy extracts the source and target IP addresses and performs intelligent encapsulation in combination with routing information to optimize the transmission path of the data packet and reduce delays caused by inaccurate routing. For SINET encapsulation, the switch fills in proprietary header information such as data priority to meet specific transmission requirements and optimize bandwidth utilization; during the decapsulation process, the protocol header is correctly removed and restored to the standard format to ensure effective forwarding of the data packet in the multi-protocol network. Through mapping lookup and protocol processing, the data transmission efficiency can be significantly improved, ensuring seamless transmission and processing of data in a multi-protocol environment, thereby effectively supporting the coordinated optimization of task offloading and data transmission in edge computing.

[0013] Step (7): The forwarding module in the switch is responsible for efficient packet reception and forwarding. Through DPDK initialization, the switch can directly access the network card hardware resources, bypassing the traditional kernel network stack, and significantly reducing the latency of packet processing. In addition, combined with CPU and thread binding technology, specific processing tasks are assigned to specified CPU cores, effectively avoiding the performance overhead caused by frequent switching of tasks between multiple cores. By optimizing the EAL parameter configuration (such as cores, ports, and queues), the switch can achieve precise resource allocation, thereby improving the parallelization capability of packet processing. On this basis, the introduction of large page memory technology reduces the fragmentation of memory allocation and reduces the cache hit failure rate, further improving the operating efficiency of the forwarding process. Through these optimization methods, the switch can achieve high-throughput, low-latency packet processing in a multi-protocol environment, meet the needs of high-performance forwarding in edge computing scenarios, and effectively improve the collaborative efficiency of data transmission and task offloading.

[0014] Step (8): For each user and edge computing node, an independent comprehensive model is established. The model defines computing power parameters and network status parameters for each edge node based on the user's location, mobile speed, computing requirements, and the processing capacity, communication delay, and service scope of the edge node, including CPU computing power, memory capacity, bandwidth resources, and network delay. Taking a typical smart city application scenario as an example, multiple sensors and user devices generate data in real time in an urban environment, such as traffic flow monitoring and environmental monitoring. In order to effectively schedule these computing tasks, it is necessary to dynamically adjust task offloading and resource allocation strategies according to changes in user location and task requirements. For example, during peak hours, the city's traffic management system needs to process a large amount of data from different sensors in real time. At this time, the edge node must reasonably schedule computing tasks and allocate processing resources to avoid excessive delays. In this process, the data quality is improved by collecting the geographic location information of users and devices, real-time computing requirements, and the load and network response of edge nodes, combined with data preprocessing techniques (such as normalization and denoising). Subsequently, a machine learning algorithm (such as support vector machine or neural network) is used to train and optimize the model, dynamically adjust the task scheduling strategy, and respond to changes in network status and user needs in real time. For example, when the network latency is high, the model can determine to offload part of the computing tasks to other edge nodes with lighter loads, or reduce computing bottlenecks through better resource allocation solutions.

[0015] Step (9): Task offloading decision. Through the improved DDPG algorithm, the network status, user location and computing requirements, as well as the load and resource conditions of the edge nodes are comprehensively evaluated to dynamically decide the target node for task offloading. The algorithm takes into account the computing power, communication delay and load conditions of the node, and generates an optimized offloading strategy, thereby effectively reducing delays and network overhead while ensuring task execution efficiency. The improved algorithm can adapt to changes in the network environment in real time, especially in scenarios with high user mobility or large fluctuations in network conditions, and continuously optimize task scheduling to ensure that tasks are completed on time. Finally, task data and configuration information are efficiently transmitted to the selected edge node to complete the computing task, promote the dynamic collaborative optimization of edge computing resources, and improve the overall system performance and resource utilization.

[0016] Step (10): Task migration decision. By continuously monitoring the user's movement trajectory, network status fluctuations and edge node load, the system can automatically trigger the task migration mechanism when it detects that the connection quality has decreased or the node load is too high. Using the improved deep deterministic policy gradient (DDPG) algorithm, the computing power, current load and communication delay of the candidate edge nodes are comprehensively analyzed to select the optimal node for task takeover. This method can effectively cope with the impact of user mobility and network condition fluctuations, and avoid delays and uneven resource allocation caused by untimely or unreasonable task migration. During the migration process, the system ensures the safe transfer of task status and data, avoids service interruption, and ensures that the task continues to be executed seamlessly on the new node. This mechanism can not only enhance the reliability of task processing, but also significantly improve the utilization efficiency and overall performance of system resources, thereby ensuring the efficiency and stability of task processing in the edge computing environment.

[0017] The technical solution of the present invention has the following advantages:

[0018] 1) Multi-protocol support and flexibility: Through the identification and preliminary processing of multiple communication protocols (such as TCP, UDP, RS232, RS485, CAN, Modbus, etc.), this method can adapt to data packets from various communication networks, improving the compatibility and flexibility of the system.

[0019] 2) Efficient protocol translation and conversion: This method ensures efficient processing and forwarding of data packets, reduces data transmission delays, and improves overall network performance through local protocol translation and conversion mechanisms and cloud-assisted strategies for handling unknown protocols.

[0020] 3) Optimized packet encapsulation and mapping lookup: Through packet tunnel encapsulation and mapping lookup strategy, this method can effectively manage and optimize the transmission path of data packets, reduce the loss and errors of data packets during transmission, and enhance the reliability of data transmission.

[0021] 4) Improved processing performance and efficiency: Through the application of DPDK technology, this method bypasses the traditional kernel network stack and directly binds to the network card, significantly improving the performance and efficiency of data processing, especially in high-speed data transmission scenarios.

[0022] 5) Dynamic and flexible task offloading and migration decisions: Using the improved deep deterministic policy gradient (DDPG) algorithm, this method can dynamically determine task offloading and migration according to the real-time network conditions and the load of the edge computing nodes, optimize the allocation of computing resources, reduce communication delays, and improve the speed and quality of task processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the specific implementation of the present invention or the technical solution in the prior art, the drawings required for use in the specific implementation or the description of the prior art will be briefly introduced below.

[0024] Figure 1 It is a flow chart of the overall solution of the present invention;

[0025] Figure 2 This is the data packet processing state transfer process of step (5) of the present invention;

[0026] Figure 3 is a schematic diagram of the task offloading and migration process of steps (8) to (10) of the present invention;

[0027] Figure 4 It is the overall algorithm flow chart of the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0029] Figure 1 is a flowchart of an optimization method for edge data transmission and computing task offloading in a computing network scenario according to an embodiment of the present invention. Figure 2 It is a state transition diagram for heterogeneous transmission protocol processing. Figure 3 This is a system architecture diagram for identity registration and resolution.

[0030] like Figure 1 As shown, the present invention comprises 10 steps.

[0031] Step (1): A protocol identification engine is set up inside the switch. The engine can automatically identify and classify multiple communication protocols including TCP, UDP, RS232, RS485, CAN, Modbus, etc. by matching the packet header. Commonly used protocol information is stored locally on the switch, and a dynamically updated protocol database is established in the cloud to store the characteristics and parsing rules of various communication protocols. The database supports administrators or automated tools to add new protocol features and update existing protocol rules, ensuring that the protocol identification engine can adapt to changes in the network environment.

[0032] Step (2): The switch has a built-in protocol conversion engine that can read the header of the data packet according to the protocol type and target forwarding port information of the data packet, and identify the protocol type of the data packet based on the specific pattern of these bytes. For protocols such as TCP, UDP, RS232, RS485, CAN, Modbus, etc., define the characteristic bytes and positions of each protocol header, and dynamically select appropriate conversion rules. Implement field extraction logic according to the specifications of each protocol. For example, for TCP and UDP data packets, extract the source port and destination port fields; for RS232 and RS485, extract the control signal of the data stream, etc. Use bit operations and byte shifting techniques to extract necessary information from the header, such as the source address, destination address, and transmission type. Write conversion rules for each known protocol, which specify how to convert the header information of the original protocol into the target protocol format.

[0033] Step (3): Establish a protocol knowledge translation database on the cloud server, which stores the characteristics, formats, and conversion rules of various known protocols. Update the database regularly to incorporate new protocol information and update the conversion rules of existing protocols. Develop an API interface for receiving packet identification requests from the switch. The interface is capable of processing query requests containing packet protocol information and transmission information. Implement request parsing logic to extract key information of the packet from the received request, such as header data, protocol type identifier, etc.

[0034] Based on the received packet information, query the protocol knowledge translation database to determine whether there is a matching protocol conversion rule. If the query is successful, extract the corresponding protocol conversion information from the database, including the new protocol header format and necessary conversion logic. Based on the query results, generate processing conversion information. This includes the new protocol header template, what information needs to be extracted from the original packet, and instructions on how to fill this information in the new header. Encapsulate the processing conversion information and send it back to the requesting switch. If an unknown protocol processing notification is received, record the relevant information and inform the switch that it cannot be processed.

[0035] Step (4): The switch parses the header fields of the received data packet to confirm whether there are local mapping rules. If so, the switch immediately constructs a new protocol header based on these rules, which includes the updated destination address and port information, so that the data packet can be efficiently forwarded according to the transport protocol. In the case where there are no local mapping rules, the switch will send a request to the cloud server to obtain the necessary mapping information. Once the mapping instructions are received from the cloud, the switch updates its local mapping table based on these instructions and adjusts the header information of the data packet accordingly to complete the tunnel encapsulation process. This process ensures that the data packet can be correctly encapsulated and forwarded to the correct destination even in the absence of local mapping information.

[0036] Step (5): The system first extracts the Ethernet header information of the data packet by parsing the Ethernet header through the parse_ethernet function. According to the value of the header field etherType, the system determines the next state: if it is 0x9999, it jumps to the identification resolution state; if it is 0x0806, it jumps to the ARP resolution state; if it is 0x0800, it jumps to the IPv4 resolution state; if it is 0x86DD, it jumps to the IPv6 resolution state; if it is any other value, it enters the default state and directly receives the data packet. The parse_arp function is used to process the ARP header information, and the data packet is directly received after extraction. The parse_sinet function is used to process the header information, and it is determined whether to jump to the IPv4 or IPv6 resolution state or enter the default state to receive the data packet according to the value of the nextheader field. The parse_ipv4 and parse_ipv6 functions are used to process the IPv4 and IPv6 header information respectively, and the data packet is directly received after extraction.

[0037] Step (6): The switch automatically identifies and processes different types of data packets according to the state migration process. First, by analyzing the target IP address of the data packet, the switch applies the ARP policy to automatically fill in the source MAC address and IP address, and sets the appropriate forwarding port. For SINET protocol data packets, the system matches according to the destination identifier, automatically sets the next hop of the data packet and updates the Ethernet frame header information. When an IPv4 or IPv6 data packet is identified and the corresponding mapping information exists, the switch prepares to perform intelligent encapsulation to optimize data transmission. In addition, during the SINET encapsulation and decapsulation process, the switch can fill in or modify the data packet header information as needed to ensure that the data packet can be correctly forwarded to the target node.

[0038] like Figure 2 As shown, after receiving the data packet, the Ethernet frame header of the data packet is first parsed to extract the MAC address of the data packet, and the next migration state is selected according to the protocol type field in the Ethernet frame; if the protocol state matches the Zhirong identification field, the Zhirong identification header is jumped to the parsing, and the data transmission destination identifier and source identifier are stored; according to the design of the identification module, the network transmission is an encapsulated data packet structure, so IPv4, IPv6 and other data packet headers need to be parsed after the identification field. At this time, the module will continue to select the parsing state according to the next header field of the data packet;

[0039] Taking IPv4 parsing as an example, after the jump state, the module stores the source IP address / destination IP address of the data packet, and continues to select the next migration state according to the protocol field in the data packet header; according to the protocol field, the data packet will jump to the TCP state or UDP state, and the information of the data packet sending port and receiving port will be stored at this time. At this point, the data packet is parsed and all data is stored for the next step of matching action selection. In addition, considering that the target IP address and MAC parsing are required for data packet transmission in the IPv4 network, the state transition diagram also includes the ARP protocol parsing state for hardware address conversion.

[0040] Step (7): First, initialize DPDK. This process is completed through the environment abstraction layer (EAL) parameter parsing, including the registration and initialization of the device driver, the acquisition of PCI network card information, and the initialization of the network card device and memory. The specific steps include:

[0041] Register and initialize the device driver, store the device driver information in the device driver information table; scan the PCI network card in the system, obtain and record the network card address, ID, device address space and driver type in the network card terminal list; establish the mapping relationship between the device driver information table and the network card terminal list, and record it in the network card driver list; map PCI resources to the virtual address space, allocate resources for the device and initialize resources; set the sending and receiving queues according to the user configuration, apply for and initialize the ring space;

[0042] After that, the data packet is transmitted in a polling-driven mode, bypassing the Linux kernel protocol stack and directly performing data packet sending and receiving processing in the user space. Specifically, it includes:

[0043] Data packet reception: Initialize the packet receiving descriptor, fill the physical address of the packet receiving buffer memory block into the descriptor, and reset the packet receiving success flag. The network card fills the received data packet into the packet receiving buffer memory block and sets the packet receiving success flag in the packet receiving descriptor. The queue polling thread checks the packet receiving success flag, stores the data packet into the function input array, and initializes the packet receiving descriptor again for the next packet receiving.

[0044] Data packet sending: Initialize the packet sending descriptor, extract the packet sending buffer memory block information and fill it into the packet sending descriptor. The network card sends the data packet according to the packet sending descriptor. The polling thread checks the packet sending end flag and recycles the packet sending descriptor and memory buffer block after completion.

[0045] Step (8): For each user and edge computing node, a comprehensive model is established. The model is based on the user's location, mobile speed, computing requirements, as well as the processing power, communication delay and service range of the edge node. Computing power parameters and network status parameters are defined for each edge node, including CPU computing power, memory capacity, bandwidth resources and network delay. Through time analysis, the time it takes for users to upload and offload tasks is calculated, taking into account factors such as channel gain and user device power. For tasks offloaded to edge computing nodes, the computing time is affected by computing power and interference power. For local computing tasks, the computing time is determined by local computing power. For high-mobility users, considering the migration of tasks, the time for migrating tasks is affected by channel gain and interference power.

[0046] Step (9): Figure 4 As shown in the figure, firstly, a state space S is established to define the state in the network environment, including network load, user request characteristics, server resource utilization and other information. The state s∈S can be expressed as s=(l, r u , c), where l is the network load, r u is the user request feature, and c is the server resource utilization. An action space A is established to define the actions that can be taken, that is, the decision to uninstall or retain a service. Action a∈A can be represented as a={0,1}, where 0 means retaining the service and 1 means uninstalling the service.

[0047] Establish a reward function R(s, a) and design a reward function to evaluate the quality of each decision. The reward function should consider factors such as resource utilization efficiency and user satisfaction. For example, when resource utilization is high and user satisfaction is improved, the reward function can give a positive reward. Specifically, the reward function can be expressed as:

[0048] R(s,a)=α×Utilization(c)+β×Satisfaction(r u )

[0049] Among them, α and β are weight coefficients, Utilization (c) represents resource utilization, and Satisfaction (r u ) indicates user satisfaction.

[0050] To build a DDPG neural network, we first build an Actor network: define a neural network, input the current state s, and output the predicted unloading strategy action a. The Actor network is used to approximate the strategy function π(s|θ π ), where θ π is the parameter of the Actor network. Establish the Critic network and define another neural network to estimate the value Q(s, a|θ) of the state-action pair Q ), where θQ is the parameter of the Critic network. The output of the Critic network can help the Actor network update its strategy. To train the DDPG algorithm, use historical data and reward signals to continuously update the parameters of the Actor and Critic networks through the DDPG algorithm so that the strategy gradually converges to the optimal strategy. The update formula of the DDPG algorithm is as follows:

[0051]

[0052] Among them, L Q is the loss function of the Critic network, J is the objective function of the Actor network, and λ Q and λ π is the learning rate.

[0053] Make strategy predictions. When the system is in a certain state, the best uninstallation strategy is predicted through the trained Actor network. This strategy can be to uninstall a service or to retain the service to maximize the cumulative reward. Make real-time decisions. According to the predicted uninstallation strategy, decide in real time whether to perform the uninstallation operation. This decision-making process can be performed periodically to adapt to changes in network status.

[0054] Step (10): The system continuously monitors the user's mobility and the status of the edge network. This includes information such as the user's bit weight, connection quality, and bandwidth utilization. Monitoring can be achieved through regular packet transmission, signal strength measurement, and network performance analysis. At the same time, the system monitors the load of edge nodes in real time, including indicators such as CPU utilization, memory usage, and task queue length.

[0055] In the task migration decision, the state of the system can be modeled as a Markov decision process (MDP). The state s∈S can be expressed as s=(p i ,q i , b i , l j , m j , t j ),in:

[0056] ·p i Is user u i The plane,

[0057] ·q i Is user u i The quality of the connection,

[0058] b i Is user u i bandwidth utilization.

[0059] ·l j is the edge node ej Load conditions (such as CPU utilization).

[0060] ·m j is the edge node e j Memory usage.

[0061] t j is the edge node e j The length of the task queue.

[0062] The action space A is defined as A = {a i j}, where a ij Is user u i Whether the task is migrated to the edge node e j Action a i j can be expressed as a ij ={0, 1}, where 0 means no migration and 1 means migration to e j .

[0063] The reward function is designed to be the negative of the total cost of task migration and task processing delay:

[0064]

[0065] Where D i is the task data size, B i j is the edge node e i and e j The communication bandwidth between them, S is the task state size, f j is the edge node e j The computing power of C 迁移 is the migration cost, which comprehensively considers the communication delay and computation delay, and α, β, and γ are weight coefficients.

[0066] In the task migration decision, the task migration decision is triggered when one of the following situations occurs: the user moves to the coverage of a new edge node; the network connection quality decreases and reaches a certain value; the edge node load is too high and reaches a certain critical value τ.

[0067] Chengtong selects the best target edge node e through the improved DDPG palm method j , to minimize the total cost of task migration and the new processing delay. The migration cost is calculated as:

[0068]

[0069] The optimization goal is to minimize the total cost of task migration and the new processing latency:

[0070]

[0071] When performing task migration, the system should save the status and data of the current task to ensure safe migration, then transfer the task data to the target edge node, start the execution of the task on the target edge node, and update the transmission path of the task status and data to ensure that users can access the migrated task.

Claims

1. An optimization method for edge data transmission and computing task offloading in a computing network scenario, characterized in that: The steps include: (1) Perform protocol identification and preliminary processing, including TCP, UDP, RS232, RS485, CAN, Modbus and other communication protocols, to determine whether the data packet is a locally known protocol; (2) Perform local protocol translation and conversion. For data packets of locally known protocols, call the protocol conversion mechanism for translation based on the protocol type and forwarding port information; (3) The cloud assists in processing unknown protocols. If the data packet is not a locally known protocol, the switch sends a query or processing request to the cloud server. After the cloud server queries the protocol knowledge translation database, it sends processing conversion information to the switch.

2. According to claim 1, the optimization method for edge data transmission and computing task offloading in a computing network scenario is characterized in that: Step (1) includes identifying and processing the data transmission destination address / source address, data transmission type and other status parameters.

3. According to claim 1, the optimization method for edge data transmission and computing task offloading in a computing network scenario is characterized in that: Step (2) includes formatting each protocol field during the protocol conversion process to ensure correct parsing and transmission of data.

4. According to claim 1, the optimization method for edge data transmission and computing task offloading in a computing network scenario is characterized in that: Step (3) includes querying the protocol definition, data structure and conversion rules related to the unknown protocol, and returning the processed conversion information.

5. An edge computing resource management system, characterized in that: The collaborative optimization method described above includes a protocol identification module, a protocol conversion module, a cloud-assisted processing module, a data packet tunnel encapsulation module, and a data packet parsing storage and state migration processing module to improve data transmission efficiency and computing task processing speed.

6. The edge computing resource management system according to claim 5, characterized in that: The protocol identification module can automatically identify data packets of various communication protocols including but not limited to TCP, UDP, RS232, RS485, CAN, Modbus, etc.

7. The edge computing resource management system according to claim 5, characterized in that: The protocol conversion module and the cloud-assisted processing module work together to process data packets of unknown protocols and optimize the data packet processing flow of known protocols.

8. A system for data packet processing and state migration, characterized in that: It includes a data packet tunnel encapsulation module and a data packet parsing storage and state migration processing module to support efficient data transmission and computing task offloading.

9. A system for data packet processing and state migration according to claim 8, characterized in that: The packet tunnel encapsulation module is based on the network transmission protocol. The switch constructs and encapsulates a new protocol header and sets a new destination, sending address and port.

10. The system for data packet processing and state migration according to claim 8, characterized in that: The data packet parsing storage and state migration processing module is responsible for the state migration processing of data packets based on the Ethernet header information and other protocol specific fields.

Images